[DRBD-user] drbd-8.3.13.tar.gz

Christian Balzer chibi at gol.com
Wed May 16 06:49:09 CEST 2012

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On Tue, 15 May 2012 17:58:53 +0200 Lars Ellenberg wrote:

> On Tue, May 15, 2012 at 11:58:57PM +0900, Christian Balzer wrote:
> > 
> > Hello Lars,
> > 
> > On Tue, 15 May 2012 15:05:50 +0200 Lars Ellenberg wrote:
> > 
> > > On Tue, May 15, 2012 at 03:24:12PM +0900, Christian Balzer wrote:
> > > > 
> > > > Hello Phillip,
> > > > 
> > > > thanks for a version that compiles out of the box on 3.x kernels.
> > > > ^o^
> > > > 
> > > > However there is one outstanding and rather critical issue not
> > > > addressed bit it, I'm talking about the lack of a:
> > > > ---
> > > > # X-Interactive:  true
> > > > ---
> > > > line in the rc.d script. 
> > > 
> > > Sorry, I'm lazy, thus asking for pointers:
> > http://wiki.debian.org/LSBInitScripts/DependencyBasedBoot
> > 
> > > And that is supposed to have what effect?
> > http://wiki.debian.org/LSBInitScripts
> > Making it actually work and not just sit there and ignore input. ^o^
> 
> It does not ignore input.
> 
Well, when did you introduce the changes below?
It definitely didn't work with a 8.3.7 (Debian) install.

I'm currently installing a test cluster (with the same flaky hardware that
caused me grief in the first time ^o^) with 8.3.13 and will test it with
and without "X-interactive" once it's done, probably sometimes next week.

> It used to expect its tty to be in "sane" mode,
> and a read on it returning a full line.
> fancy boot splash thingies tend to put the tty in raw mode,
> which we now handle as well (input comes in one byte a time).
> 
> Sometimes said fancy boot thingies even leave the tty connected
> to tty1 (say) and present you tty7 (or so).
> in which case you'd need to first (CTRL?-) ALT-F1 so the input
> gets where it belongs ;-)
> 
> Which is why we now do
>  exec > /dev/null 2>&1; plymouth quit ; usplash_write QUIT ;
>  stty echo icanon icrnl
> before we wait for user input in that case...
> 
> Still, if that magic init script header is supposed to take care of it,
> I'm very much pleased to have learned that lesson ;)
> 
> > For example another init script in Debian testing marked interactive
> > is apache2, for those really paranoid people who have passwords on
> > their SSL keys. ^.^
> > 
> > > And that is supposed to work with which distributions?
> > > And that is documented where?
> > > 
> > See first URL, it being LSB pretty much every distro. 
> 
> Glad I put "supposed to" in that question ;-)
> 
> > 
> > > > This was recently fixed in Debian and has bitten yours truly quite
> > > > painfully in the ass twice, before I realized what was going on.
> > > > 
> > > > Since "wfc-timeout" defaults to 0 and pretty much all example
> > > > configurations suggest that that as well the init script really
> > > > needs to interactive. 
> > > > 
> > > > I experienced a double (both nodes) crash in the past, one node
> > > > never recovered and the other was stuck in the boot process
> > > > waiting for it with no way to resume booting due to the inability
> > > > to actually enter "yes" at the prompt...
> > > 
> > > The "yes" should work now, btw,
> > > and you can always ssh in and kill the drbdsetup wait-conn processes.
> > > 
> > I don't think it works on systems with dependency based boot w/o that
> > X-Interactive line. And I think it hangs in the boot process before
> > sshd is started (though I really haven't tested and thought of that),
> > certainly before another console screen and logins are available.
> 
> We have # Should-Start:   sshd 
> so if this is supposed to check on X-Interactive, it should well be able
> to act on Should-Start
> 
> Ok, so your sshd is not "sshd".
> Tough luck :-/
> 
Again, I didn't even think of checking for connectivity via SSH at that
time. Also a running sshd might still be defeated by a
"DELAYLOGIN=yes" (setting and not removing /etc/nologin until everything
is done), though "no" is the default in Debian at least.

> > > > Would be nice if this could be fixed for any future releases and
> > > > heads-up to anybody reading this.
> > > 
> > > If we knew about that particular magic,
> > > we might even have added that line already.
> > > 
> > > But sorry, I don't really read (or follow)
> > > init script magic comment RFCs too often ;-)
> > > 
> > > Thanks for bringing this up.
> > > 
> > No problem, I thought it might have caught your eyes when first doing
> > that LSB header or when it was reported and fixed in Debian.
> > 
> > > If you had brought it up during RC,
> > > you probably had it in now.
> > > 
> > See above, I was fully expecting it to be there and had no time for RC
> > testing. Oh well, hopefully people will read this until the next
> > release.
> 
> http://git.drbd.org/gitweb.cgi?p=drbd-8.3.git;a=commitdiff;h=2c5b35d
> 
Thanks!

Christian
-- 
Christian Balzer        Network/Systems Engineer                
chibi at gol.com   	Global OnLine Japan/Fusion Communications
http://www.gol.com/



More information about the drbd-user mailing list